성능 데이터 분석

Welcome! In this lesson, we'll dive into the art of analyzing the performance data collected by Real User Monitoring (RUM) and Synthetic Monitoring tools.

Understanding this data is crucial for identifying trends, pinpointing issues, and ultimately making your website faster for everyone.

RUM tools gather data directly from your users' browsers. This includes a wealth of metrics like:

• Page Load Time: Total time for a page to fully load.
• Core Web Vitals: LCP, FID, CLS – key user experience metrics.
• Resource Timings: How long specific images, scripts, or stylesheets take to load.
• JavaScript Errors: Client-side errors impacting user experience.

RUM shows you what users actually experience.

When looking at RUM data, simple averages can be misleading. Instead, focus on percentiles.

• Median (50th percentile): Half of your users experience this or better.
• 90th Percentile: 90% of users experience this or better. This is often a good target for optimization.
• 99th Percentile: The experience of your slowest users. Addressing this can uncover critical issues.

A high 90th percentile LCP means many users are waiting too long.

Analyzing data over time helps identify trends and regressions. A trend might be a gradual improvement from optimizations, or a slow degradation due to new features.

Regressions are sudden drops in performance, often after a new deployment. Dashboards typically show graphs of metrics over days or weeks, making these changes visible.

• Upward spike: Performance got worse.
• Downward dip: Performance improved.

Synthetic monitoring runs automated tests from controlled environments. This means consistent network speeds, device types, and locations.

Key synthetic metrics include:

• Speed Index: How quickly content visually populates.
• Time to First Byte (TTFB): Time until the server responds.
• Lighthouse Scores: A comprehensive audit score for performance, accessibility, etc.

Synthetic data is excellent for catching regressions in a controlled, repeatable way.

RUM and Synthetic data offer different, yet complementary, views.

• RUM: Shows the messy, real-world experience (varied networks, devices, locations). Great for understanding user impact.
• Synthetic: Provides a baseline, repeatable performance measurement. Ideal for CI/CD and catching regressions early.

Using both gives you a holistic understanding of your site's performance.

Performance isn't just a technical metric; it directly impacts your business. Analyzing data helps you connect the dots:

• Conversion Rates: Faster sites often lead to higher sales or sign-ups.
• Bounce Rate: Slow pages make users leave quickly.
• User Engagement: Responsive sites keep users interacting longer.

Quantify the impact of performance improvements to justify further investment.

Once you spot an issue, you need to drill down. Most monitoring tools allow you to filter data by:

• Page URL: Is the issue specific to one page?
• Browser/Device: Does it affect only mobile users or a specific browser?
• Geographic Location: Are users in certain regions experiencing slowness?
• Resource Type: Is it a large image, a slow API call, or a blocking script?

This helps narrow down the problem area for your team.

Don't wait for users to report issues. Set up alerts to be notified automatically when performance metrics cross critical thresholds.

Common alert conditions include:

• 90th percentile LCP exceeds X seconds.
• Cumulative Layout Shift (CLS) score is above Y.
• Error rate on a critical page increases by Z%.

This allows for quick response and minimizes user impact.

The goal of data analysis isn't just to find problems, but to fix them. Prioritize optimizations based on:

• Impact: How many users are affected? How critical is the page/feature?
• Effort: How difficult is the fix?
• Business Value: What's the potential ROI?

Start with high-impact, low-effort changes for quick wins, then tackle bigger projects.

Your team just deployed a new feature. Looking at your RUM dashboard, you notice the 90th percentile LCP for mobile users on the homepage jumped from 2.5 seconds to 5 seconds. The median LCP for desktop users remained stable at 1.8 seconds.

What are the most likely immediate conclusions you can draw from this data?

Great job! You've learned how to effectively analyze RUM and Synthetic monitoring data.

Remember to look beyond averages, track trends, correlate performance with business outcomes, and drill down to find root causes. Proactive alerting and turning data into actionable insights are key to continuous optimization.

Keep monitoring, keep optimizing!